Re: [PATCH] memory ordering in __kfifo primitives

["Paul E. McKenney" <paulmck@us.ibm.com>]

On Thu, Aug 10, 2006 at 10:27:42PM +0200, Stelian Pop wrote:
> [open-iscsi@googlegroups.com bouncing, removed from CC:]
> 
> Le jeudi 10 août 2006 à 09:47 -0700, Paul E. McKenney a écrit :
> 
> > > Let's take this problem differently: is a memory barrier cheaper than a
> > > spinlock ? 
> > 
> > Almost always, yes.  But a spinlock is cheaper than a spinlock plus
> > a pair of memory barriers.
> 
> Right, but I think we're optimizing too much here. 

That was in fact my point initially -- why not just require locking,
either that registered at kfifo_alloc() time or a separately acquired
lock?

> > > If the answer is yes as I suspect, why should the kfifo API force the
> > > user to take a spinlock ?
> > 
> > My concern is that currently a majority of the calls to __kfifo_{get,put}()
> > are already holding a spinlock.
> > 
> > But if you could send me your tests for lock-free __kfifo_{get,put}(),
> > I would be happy to run them on weak-memory-consistency model machines
> > with the memory barriers.  And without the memory barriers -- we need
> > a test that fails in the latter case to prove that the memory barriers
> > really are in the right place and that all of them are present.
> > 
> > Does this sound reasonable?
> 
> It would sound reasonable if I had any tests to send to you :)
> 
> Since I don't have any and since you're the one proposing the change, I
> guess it's up to you to write them. :)

Ah, but you owe a test debt from your earlier submission of kfifo!  ;-)

						Thanx, Paul